Research Institute of Industrial Economics

P.O. Box 55665

SE-102 15 Stockholm, Sweden

info@ifn.se

www.ifn.se

IFN Working Paper No. 954, 2013

Hierarchies, the Small Firm Effect, and Entrepreneurship: Evidence from Swedish Microdata

Joacim Tåg, Thomas Åstebro and Peter Thompson

Hierarchies, the Small Firm Effect, and Entrepreneurship:

Evidence from Swedish Microdata

Joacim Tåg*

Research Institute of Industrial Economics (IFN)

Thomas Åstebro**

HEC Paris

Peter Thompson***

Emory University

February 2013

We explore whether the tendency for smaller firms to have fewer hierarchical layers ex-

plains the well-documented inverse correlation between firm size and the rate at which

employees become business owners. Our analysis is based on a Swedish matched employ-

er-employee dataset. Conditional on firm size, employees in firms with more layers are

less likely to enter entrepreneurship, to become self-employed, and to switch to another

employer. The effects of layers are much stronger for business creation than for job-

switching and they are stronger for entrepreneurship than for self-employment. Howev-

er, hierarchies constitute only a partial explanation of the small firm effect. Potential

explanations for the effects of layers are examined. Part of the effect appears to be due to

preference sorting by employees, and part due to employees in firms with fewer layers

having a broader range of skills.

Key words: entrepreneurship; employee mobility; hierarchy, rank, small firm effect.

JEL codes: L26, D20, J20, M50.

*Research Institute of Industrial Economics (IFN), Box 55665, SE-102 15 Stockholm, Sweden. E-

mail: joacim.tag@ifn.se. Joacim gratefully acknowledges financial support from Vinnova and the

Jan Wallander and Tom Hedelius Foundation. **HEC Paris, 1 rue de la Liberation, 78351 Jouy-

en-Josas Cedex, France. Email: astebro@hec.fr. ***Goizueta Business School, Emory University,

1300 Clifton Road, Atlanta, GA 30322, USA. Email: peter.thompson@emory.edu.

1

1. Introduction

Compared with their counterparts in large firms, workers in small firms are more like-

ly to separate from their employers (Anderson and Meyer, 1994; Lazear and Shaw,

2008) and those that leave small firms are more likely to become entrepreneurs or

self-employed than those leaving large firms (Wagner, 2004; Dobrev and Barnett,

2005; Gompers et al., 2005; Elfenbein, Hamilton and Zenger, 2010; Chen, 2012). While

higher job separation rates in small firms are in part due to the greater volatility of

jobs and job types in small and young firms (Haltiwanger, Jarmin and Miranda, 2011),

higher rates of entrepreneurship among movers suggests that employees of small

firms are different from their counterparts in large firms. We will follow Elfenbein et

al. (2010) and refer to the greater propensity of employees of small firms to establish

businesses as the small firm effect.

One potential explanation for the small firm effect is the segregation of observable

worker types in to large and small firms. For example, Frederiksen (2006) shows that

women in Denmark are both more likely to separate from their jobs and more likely to

be employed in a small firm. Similarly, separation rates are higher for the less educated

and educational attainment is on average lower in small firms. Because small firms are

younger on average, job tenure is lower and this, too, is associated with higher separa-

tion rates. But these readily observable characteristics of employees in small firms do

not explain higher rates of entrepreneurship: women are markedly less likely to be-

come entrepreneurs than men, and neither tenure nor education are good predictors

of entrepreneurship among movers.

Elfenbein et al. (2010) find evidence of segregation of unobservable worker types.

First, small firms offer greater autonomy and a greater variety of work experiences, so

we might expect that workers with a preference for these work attributes are more

likely to be employed by, and to create, small firms. Elfenbein et al. have a direct meas-

ure of whether individuals in their sample have a desire to become entrepreneurs.

Individuals who do are overrepresented in small firms, and are more likely to subse-

quently establish a business. Second, employees in small firms may have skills that are

especially valuable for the creation of new businesses. These skills may reflect innate

abilities that are better rewarded in small firms so future entrepreneurs select into

small firms, or they may be abilities that are gained as a result of employment in small

firms. Consistent with this hypothesis, Elfenbein et al. find that entrepreneurs with

prior small firm experience have better performance.

Small firms attract individuals with preferences and abilities for entrepreneurship not

just because they are smaller, but because they are different. In particular, small firms

are less bureaucratic, and bureaucracy stifles the accumulation of skills appropriate

for entrepreneurship and repels those with a preference for autonomy and work varie-

ty. Sørensen (2007) reviews the prior literature and sociological arguments for the

negative effects of bureaucracy on entrepreneurship, and examines their implications

2

using Danish microdata. However, lacking direct measures of the extent of bureaucra-

cy, Sørensen associates greater levels of bureaucracy with older and larger firms. As a

result, one cannot infer from his analysis the extent to which varying levels of bureau-

cracy can explain the small firm effect.

In this paper, we use the number of management layers in a firm as an indicator for the

extent of bureaucracy that is independent of firm size, and investigate the extent to

which the correlation between firm size and its hierarchical structure can explain the

small firm effect. Our sample is drawn from Swedish matched employer-employee

dataset, and consists of more than 100,000 firm-level observations and 7 million indi-

vidual-level observations over the period 2001–2008. As our discussion of the data in

Section 2 will make clear, our sample is not a true panel although we do have repeated

observations on both individuals and firms. Following Caliendo and Rossi-Hansberg

(2012) and Caliendo, Monte and Rossi-Hansberg (2012), we identify employees’ ranks

from their job titles and measure the hierarchal structure of the firm by the number of

layers of management it has. Our sample behaves as one would expect from a mean-

ingful measure of hierarchy: upper levels of management contain fewer employees and

pay higher wages than lower levels, and employee transitions are most likely to be to

an adjacent rank.

In Section 3, we demonstrate that a strong small firm effect exists in Sweden, and that

there is a strong correlation between the number of management layers in a firm and

its size. We also show that, conditional on size, employees in firms with more man-

agement layers are less mobile. This is true regardless of the mover’s destination, but

the effect of layers on business creation is much greater than its effect on relocation to

other incumbent firms. We decompose business creation into self-employment and

entrepreneurship, defined by the legal form of business that is created. More layers in

a firm suppress both types of business creation. However, despite the magnitude of the

effect of layers, the small firm effect persists. Our point estimates suggest that the hier-

archical structure of the firm accounts for about one fifth of the small firm effect.

In Section 4, we examine potential explanations for the effect of layers on mobility in

general and business creation in particular. We first look at evidence for the prefer-

ence sorting hypothesis. Individuals with a strong preference for entrepreneurship

should be more willing to establish businesses that yield low financial returns, and

they should also persist in business for longer. We test these implications, and con-

clude that employees of small firms do behave in ways consistent with preference sort-

ing. We then assess whether there are differences in entrepreneurial ability among

employees of firms with different layers of management. We frame our analysis

around Caliendo and Rossi-Hansberg’s information-processing model, which relates

the number of management layers in a firm to the breadth of problems that employees

at different ranks are expected to be able to solve. Although they do not study their

model’s implications for employee separations, we derive and test two implications.

3

First, employees in firms of a given size but with fewer layers should have a higher

propensity to enter entrepreneurship and self-employment and to be more successful

when they do. Our evidence on business creation rates are consistent with the infor-

mation-processing model (but also with preference sorting); our evidence on earnings

do not support the model. Second, the propensity for business creation is greater when

employees of a given rank have more layers beneath them. We find string support for

this prediction.

Our third investigation is an examination of the career concerns model. Hierarchies

offer opportunities for promotion and if the increment to wages that is associated with

a rise in rank is large, the opportunity cost of mobility increases and job separation

rates will decline. We consider three distinct measures of a firm’s wage policy, but find

no evidence to support the career concerns hypothesis. Finally, we explore whether

the model of job mismatching developed in Åstebro, Chen and Thompson (2011) can

explain the effect of layers. In their model, mismatching occurs when employees of

differing skills are put to work together. Such mismatching is more likely in the tails of

the ability distribution so we see greater mobility from the tails. We hypothesize that

layers create a functional separation between agents of high and low ability, and

thereby reduces mismatching problems. We find clear evidence that such mismatching

is occurring in our sample, but no evidence that this explains the effect of layers.

2. The Data

The data are drawn from two distinct sources, one at the firm level and one at the indi-

vidual level. The firm level dataset comes from the IFN Corporate Database (IFNCDB),

which is based on official mandatory accounting data filed to the Swedish Companies

Registration Office. The individual level dataset is based on the Statistics Sweden’s

LISA database drawing on several official registry databases of every person living in

Sweden. From the IFNCDB we extract the annual accounts of firms, and from the LISA

database we obtain information on occupation codes, firm-worker links, worker’s la-

bor income, worker’s capital income, and numerous other worker characteristics. We

make use of firm- and individual-level data for the period 2001–2008.1 We drop dupli-

cated firm-year information (multiple annual accounts can be submitted each year)

and then merge the firm and individual level data. We drop firms active in the health,

education, agriculture and fishing industries, and also firms in the public sector, in

order to focus on private sector firms. Given our focus on hierarchies, we restrict at-

tention to firms with more than 5 employees (we lose 76 percent of all firm-year ob-

servations since most Swedish firms are small). We also restrict attention to individu-

als between the ages of 20 and 60.

1 Accurate occupation data is only available from 2001 and onwards.

4

Because we will construct indicators of employee rank and organizational structure

from information on occupational codes assigned by Statistics Sweden, we further re-

strict our sample to include only firms with sufficient occupation data for its employ-

ees. In the sample of firms with more than 5 employees, 63 percent of all individual-

level observations have accurate occupation data as the occupation data is based on

surveys and does not have complete coverage. Because occupation data is not com-

plete for every year, we drop firm-year observations with fewer than 75 percent of all

employees having accurate occupation data. This involves mainly smaller firms that

are less likely to be sampled by Statistics Sweden and it reduces the firm-year sample

size by another 76 percent. In our final sample, 91 percent of all individual-level ob-

servations have accurate occupation data and it covers 48 percent of all individual-

year observations in the initial dataset.

In each year, we track whether employees remained with their current firm, switched

to another incumbent firm, or created and became primarily occupied in running their

own business. We follow the definition used by Statistics Sweden to define entrepre-

neurs and self-employed. Statistics Sweden defines an individual as being employed in

her own firm in a given year if her total income from her own company (labor and cap-

ital income) is greater than 62.5 percent of all other labor income.2 We define an indi-

vidual as entering entrepreneurship in any given year if the following criteria are sim-

ultaneously fulfilled:

1. Newly occupied in own business. An individual is classified by Statistics Swe-

den as working in her own company in the current year, but had not been in the

previous year.

2. New place of work. The individual’s current firm and establishment identifiers

are different from the previous year, and

3. New firm. No individual in our sample had worked for the current firm in the

previous year.

Statistics Sweden also separates sole proprietorships from limited liability companies.

Because an individual who intends to create a growing company likely will establish a

limited liability company, we characterize agents starting limited liability companies

as “entrepreneurs”, and agents starting sole proprietorships “self-employed”.3

2 Statistics Sweden treats self-employment as the primary occupation even though it may gen-erate less income than other activities, because self-employment typically generates lower hourly wages.

3 The mean size of a limited liability company two years after creation is 5.3 employees with a standard deviation of 8.3, while the mean size of a sole proprietorship is 1.2 employees with a standard deviation of 0.9. Henrekson and Sanandaji (2013) survey the literature on entrepre-neurship and self-employment and demonstrate the importance of separating between entre-preneurs and the self-employed using data on billionaire entrepreneurs.

5

Table 1

Sample size

Year Firms Workers To other firm To self

employment To

entrepreneurship

2001 20,364 1,042,267 155,063 3,061 465

2002 19,051 1,155,587 156,242 3,451 434

2003 13,535 1,079,013 139,370 4,103 1,125

2004 11,831 1,048,657 137,992 4,249 1,103

2005 12,426 1,059,188 140,544 4,329 1,042

2006 13,026 1,060,428 146,305 3,763 1,268

2007 15,516 1,107,941 152,545 4,050 1,537

Total 105,749 7,553,081 1,028,061 27,006 6,974

Table 1 displays numbers on the sample size of the final dataset. It contains 105,749

firm-year observations covering 61 percent of value added and 53 percent of employ-

ment in the Swedish private sector. Of around 7.5 million employee-level observations,

about 14 percent transitions to a new incumbent employer in each year, about 0.36

percent enter self-employment, while just 0.1 percent enters entrepreneurship. The

number of firms in the sample varies considerably across years, likely a result of dif-

ferent sampling of occupation codes over time. Moreover, parts of the surveys are tar-

geted to certain sectors in certain years, which probably accounts for the large jumps

in the number of observations per year.

2.1 Occupation classifications, rank and management layers

Our data on the hierarchical structure of firms are developed out of occupational clas-

sifications obtained from Statistics Sweden’s LISA database. The Swedish Standard

Classification of Occupations 1996 (SSYK) is a Swedish version of the International

Standard Classification of Occupations (ISCO-88). The SSYK data available from the

LISA database come mainly from the official wage statistics survey (Lönestruktur-

statistiken) and from a supplementary survey of firms not included in the official wage

survey. Between the two surveys, at least 40,000 firms are sampled every year. The

sampling design is a rolling panel,4 and all eligible firms are surveyed at least once eve-

ry five years. Not all firms are included in these surveys. The largest excluded category

is of self-employed workers who do not obtain any wage from a limited liability com-

pany, but the surveys also exclude owners who receive payment from their companies

only in the form of dividends.5

4 Except that firms with at least 500 employees are always included.

5 There have been some attempts since 2004 to survey partnerships and sole proprietorships

6

Table 2

Occupational classifications, skill levels, and rank

SYYK Occupational Classification Skill Level Rank

1. Legislators, senior officials & managers NA 3. CEOs: SSYK 121 (Directors and chief

executives), 131 (Managers of small en-

terprises), 111 (legislators and senior

government officials), 112 (senior officials

of special-interest organizations)

2. Senior staff: SSYK 122 (Production and

operations managers), 123 (Other special-

ist managers)

2. Professionals 4 1. Supervisors: SSYK 200-399 (Profes-

sionals, technicians and associate profes-

sionals)

3. Technicians & associate professionals 3

4. Clerks 2 0. Production workers: SSYK 400-999

(Clerks, service workers and shop sales

workers, skilled agricultural and fishery

workers, craft and related trades workers,

plant and machine operators and assem-

blers, and elementary occupations).

5. Service workers & shop sales workers 2

6. Skilled agricultural & fishery workers 2

7. Craft & related trades workers 2

8. Plant & machine operators & assemblers 2

9. Elementary occupations 1

0. Armed forces NA Omitted

The SSYK assigns workers to one of ten main occupational categories, and one of a

large number of subcategories. Statistics Sweden also assigns each of these occupa-

tional categories to one of four skill levels.6 We use the SSYK codes to assign a rank to

each employee (see Table 2). The highest rank, which we label CEOs, consists of direc-

tors, chief executives, managers of small enterprises, and certain other senior officials.

The next two ranks comprise two levels of managements. The more senior, which we

label as “senior staff”, contain production and operations managers and certain other

specialist managers. The less senior, “supervisors”, consists of workers with occupa-

tions classified in the SSYK as professionals, technicians and associate professionals.

The fourth category, “production workers”, comprises clerks, service workers, plant

and machine operators, and other non-supervisory positions. As Table 2 shows, our

but the coverage is extremely limited.

6 Statistics Sweden notes that, although these skill levels have been made operational in terms of the educational categories of the International Standard Classification of Education, they do not imply that the skills necessary to perform the tasks and duties of a given job can be ac-quired only through formal education.

7

rank classifications broadly coincide with the skill levels attributed to these positions

by Statistics Sweden.

Do our classifications induce patterns consistent with our notions of rank and hierar-

chies? We look first at earnings by rank. The earnings data we use throughout the pa-

per measure each individual’s annual labor income. However, for compactness, we will

generally refer to our earnings measure as the “wage”. Table 3 displays the wage dis-

tribution by rank. Clearly, workers in higher ranks tend to earn more on average than

workers in lower ranks. The same ordering persists at each percentile shown in the

table, with the notable exception of senior staff that earn more than the CEO rank at

the lower percentiles. This reflect in part differences across firms (large firms that pay

more on average are more likely to have senior staff) and in part the tendency of sen-

ior staff to be highly educated. The table also displays that the within-rank wage dis-

tribution is the highest at the top rank and reduces monotonically with lower rank.

Table 4 plots rank transitions (of course the lowest and highest ranks can only transi-

tion in one direction) among workers remaining with the same firm. Transition rates

decline with distance between rank pairs, and most employees do not change rank. For

example, 93.6 percent of production workers are at the same rank three years later,

while 5.4 percent are at the rank of supervisor and less than one percent have attained

either of the two highest ranks. Similarly CEOs are most likely to remain as CEOs;

among those that do switch rank, moving one rank is almost twice as likely as moving

two, and moving two is in turn twice as likely as moving three. Tables 3 and 4 together

suggest that our occupational classifications succeed in capturing a form of distance

consistent with our notion of ranks and hierarchies.

Table 3

Wage distribution across broad occupation classes based on SSYK

Rank Mean Percentiles Wage

Dispersion 10th 25th 50th 75th 90th

3. CEOs and directors 600 232 300 417 669 1,086 1,324

2. Senior staff 540 292 363 469 620 833 775

1. Supervisors 358 201 260 330 421 534 475

0. Production workers 239 127 192 241 288 336 277

Mean 302 151 215 272 348 462 388

Note: Data are in units of 1,000 SEK in 2005 prices.

8

Table 4

Rates of within-company rank transitions

To production

worker

To

supervisor

To

senior staff To CEO

Production worker 93.61 5.44 0.71 0.24

Supervisors 7.63 87.32 4.34 0.71

Senior Staff 4.54 25.55 63.24 7.67

CEO or director 7.10 13.59 22.20 57.11

Transition rates over a three-year period if available; otherwise two years, otherwise one year. Transition rates are very similar if we include employees that switched firms.

Table 5 provides information about within-firm differences in rank populations and in

earnings across ranks. Following Caliendo et al. (2012), a firm-year observation with R

ranks will be said to have L=R s

the probabilities that a lower rank contains more employees than the rank immediate-

ly above it in the firm. In the large majority of cases, the firm’s structure corresponds

to our notion of hierarchies as triangular structures, where higher ranks consist of

small numbers of people supervising larger groups of workers in lower ranks. For ex-

ample, in firms with only one layer of management, the lower of the two ranks in the

firm has more workers in 88 percent of the firm-year observations. Panel B compares

within-firm earnings across ranks. It shows that members of the higher rank earn

more on average than workers in the rank immediately below them in the firm. For

example, in firms with three layers of management, the lowest rank has lower mean

earnings than the next layer in 94 percent of the firm-year observations. These per-

centages are consistent with the earnings distributions summarized in Table 3.

Table 5

Size and wage by layers of management

Layers of management

A

Size

B

Labor Income

0>1 1>2 2>3 0

9

3. Hierarchies and the Small Firm Effect

In this section we explore the relationship between worker mobility, especially into

entrepreneurship and self-employment, and the hierarchical structure of the firm, and

we assess whether the hierarchies offer an explanation for the small firm effect. Table

6, which summarize the raw data on management layers, firm size and earnings, re-

veals patterns consistent with the discussion in section 1. First, firms with more layers

of management tend to be larger, whether size is measured by value added or by the

number of employees. For example, firms with three layers of management have on

average 15 times as many employees as firms with just one layer, and they also pro-

duce 38 percent more value added per worker. Table 6 also documents that firms with

more layers of management pay higher wages. Finally, firms with more layers of man-

agement also have a larger dispersion of wages. If more layers of management, higher

wages and higher wage dispersion all suppress entry into entrepreneurship and self-

employment, then these positive associations between layers, firm size and wages

might well enable hierarchies to explain the firm-size effect.

Table 7 reports the main results of this section. We estimate multinomial logit regres-

sions with four possible outcomes in each year: remain with the current employer,

enter entrepreneurship, enter self-employment, and switch to another employer.

Model A, in columns (1) through (3), includes indicators for firm size, while model B in

the remaining columns adds controls for the number of management layers. The key

result here is that an increase in the number of layers is negatively associated with

mobility of all types, although the effect is much stronger for entry into entrepreneur-

ship and self-employment than it is for moving to another employer. However, the

inclusion of controls for layers has only a modest effect on the estimated impact of firm

size on entry into entrepreneurship and self-employment.

Table 6

Descriptive statistics of sample by layers of management in firm

Layers of

management N

Means Median

Wage Value Added

per worker Employees Wage

Wage

Dispersion

0 21,627 477 13 235 317 225

1 35,245 594 18 253 383 237

2 28,773 717 62 288 457 268

3 20,104 817 268 319 468 295

Total 105,749 646 77 271 406 254

N is the number of firm-year observations. Value added per worker, annual wages and wage

dispersion are in unit of 1,000 SEK at 2005 prices

10

Table 7 Worker Transitions. Multinomial Logit Estimation

Model A Model B

Entrepre-neurship

Self-employment

Other firm Entrepre-neurship

Self-employment

Other firm

(1) (2) (3) (4) (5) (6) CEOs and Directors 1.054*** 0.010 -0.034*** 1.124*** 0.046 -0.030** (0.067) (0.060) (0.012) (0.068) (0.060) (0.012) Senior staff 0.994*** 0.400*** 0.278*** 1.048*** 0.419*** 0.282*** (0.054) (0.036) (0.007) (0.054) (0.037) (0.007) Supervisors 0.748*** 0.259*** 0.031*** 0.776*** 0.269*** 0.033*** (0.039) (0.020) (0.003) (0.039) (0.020) (0.003) Firm layers: 1 -0.250*** -0.200*** -0.042*** (0.061) (0.032) (0.007) Firm layers: 2 -0.336*** -0.243*** -0.078*** (0.060) (0.032) (0.007) Firm layers: 3 -0.461*** -0.215*** -0.055*** (0.063) (0.034) (0.007) Size 50-100 -0.470*** -0.388*** -0.063*** -0.379*** -0.350*** -0.052*** (0.047) (0.026) (0.005) (0.049) (0.027) (0.005) Size 100-500 -0.770*** -0.661*** -0.136*** -0.645*** -0.616*** -0.123*** (0.038) (0.021) (0.004) (0.042) (0.024) (0.004) Size 500-1500 -0.831*** -0.851*** -0.255*** -0.684*** -0.808*** -0.243*** (0.044) (0.025) (0.004) (0.049) (0.029) (0.005) Size >1500 -1.281*** -1.172*** -0.579*** -1.120*** -1.123*** -0.567*** (0.043) (0.023) (0.004) (0.050) (0.028) (0.005) Age (years) 0.152*** 0.107*** -0.080*** 0.152*** 0.107*** -0.080*** (0.012) (0.006) (0.001) (0.012) (0.006) (0.001) Age squared -0.002*** -0.001*** 0.001*** -0.002*** -0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female -0.855*** -0.653*** -0.094*** -0.850*** -0.650*** -0.093*** (0.039) (0.019) (0.003) (0.039) (0.019) (0.003) Education 0.046*** 0.095*** 0.105*** 0.046*** 0.095*** 0.105*** (0.015) (0.008) (0.001) (0.015) (0.008) (0.001) Log(wage) 0.496*** -0.768*** -0.531*** 0.499*** -0.766*** -0.530*** (0.035) (0.011) (0.002) (0.035) (0.011) (0.002) Tenure (years) -0.086*** -0.134*** -0.156*** -0.085*** -0.134*** -0.156*** (0.010) (0.006) (0.001) (0.010) (0.006) (0.001) Tenure squared 0.001* 0.004*** 0.005*** 0.001* 0.004*** 0.005*** (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) Constant -12.876*** -3.024*** 3.253*** -12.624*** -2.862*** 3.295*** (0.298) (0.128) (0.021) (0.301) (0.130) (0.022) N 6,865,026 6,865,026 6,865,026 6,865,026 6,865,026 6,865,026

Standard errors in parentheses. . *** p

11

ated with mobility of all kinds,7 while the more educated8 and males are more mobile

regardless of destination. Interestingly, individual wages have impacts that differ by

destination. Individuals with higher wages are less likely to switch to another incum-

bent employer and less likely to enter self-employment. However, a higher wage is

associated with an increase in the propensity to become an entrepreneur. The con-

trasts between these effects of individual wages are not only statistically significant,

they are economically meaningful. For example, a one standard deviation increase in

log income is associated with a decline in the odds of switching employers (relative to

staying with the current employer) of 15 percent, with a decline in the odds of entering

self-employment of 38 percent, but with an increase in the odds of entering entrepre-

neurship of 4.4 percent. Employee age similarly has disparate effects on mobility by

destination. Increasing age raises the probability of entering entrepreneurship and

self-employment until a peak hazard is attained at 43 years of age for entrepreneur-

ship and 48 years for self-employment, after which the hazard declines. In contrast,

increasing age reduces the probability of switching incumbent employers until a mini-

mum hazard is attained at about 47 years of age.

These results for the control variables suggest that mobility is driven by a complex

interplay of multiple forces. For examples: the negative effect of tenure on mobility of

all kinds is consistent both with job matching models (e.g., Jovanovic, 1979) and with

survivor bias caused by unobserved heterogeneity across individuals in the propensity

to move; the initially positive impact of age on business creation might reflect the ef-

fect of wealth constraints that are relaxed as an agent ages and saves, or the conse-

quences of on-the-job learning specific to the demands of business creation; and the

contrast between the effects of individual wages on entrepreneurship and self-

employment recalls the mismatching model of Åstebro, Chen and Thompson (2011), in

which business creation is more likely among agents with especially high and especial-

ly low ability.

We return now to consideration of the main effects of firm size, rank, and management

layers on mobility. Consider first the role of rank. Employees in supervisory positions

are much more likely than production workers (the omitted category) to enter entre-

preneurship, and the propensity to do so rises with each increase in rank. CEOs, direc-

7 The positive coefficients on the quadratic terms indicate a non-monotonic effect of tenure. For entry into entrepreneurship and self-employment the minima are attained at 39 and 18 years of tenure; for movements to another incumbent employer the minimum is at 15 years. Howev-er, the predicted effect of tenure is greater than these minima and extrapolates outside most of our sample observations.

8 Education is on a scale from 1–6 corresponding to: 6. Postgraduate education; 5. Post-secondary education, two years or longer; 4. Post-secondary education, less than two years; 3. Upper secondary education; 2. Primary and lower secondary education; 9 or 10 years; and 1. Primary and lower secondary education, less than 9 years.

12

tors and senior staff are almost three times more likely than production workers to

found a limited liability company, while supervisors are more than twice as likely to do

so (Figure 1 plots the odds ratios). Senior staff and workers with supervisory rank are

also more likely to enter self-employment. The effects are smaller than for entrepre-

neurship, however, and CEOs are no more likely to become self-employed than pro-

duction workers.9 Finally, CEOs and Directors are less likely to switch employers than

production workers, while Senior Staff and Supervisors are more likely to do so than

production workers. In this case, however, the magnitudes of the differences are trivi-

al.

Table 7 contains four indicators for size categories, the omitted category being firms

with fewer than 50 employees. The point estimates reveal a strong negative effect of

increasing firm size on mobility; regardless of destination (see also Figure 2). For ex-

ample, employees in the largest firms are only one third as likely as employees in the

smallest firms to enter entrepreneurship or self-employment, and they are 44 percent

less likely to switch to another employer. The effect of firm size is not limited to a con-

trast between the largest and smallest firms. As Figure 2 illustrates, each change in

firm size class is in almost all cases associated with a similar change in the odds of mo-

bility. Clearly, there is a small firm effect on mobility in general. However, the differen-

tial effect of firm size on movements into business creation relative to movement to

other firms also demonstrates that there is a strong small firm effect on entrepreneur-

ship and self-employment even conditional on job separation. Model B in Table 7 in-

cludes controls for layers of management and shows that, even though we condition on

firm size and employee rank, more layers are associated with less mobility of all kinds

(see Figure 3 for the odds ratios). Business creation rates, especially in entrepreneur-

ship, are strongly affected by adding layers of management. For example, moving from

zero to one layer of management reduces the likelihood of a transition to entrepre-

neurship by 22 percent and the likelihood of a transition to self-employment by 18

percent. There is, however, an interesting contrast between the effects of layers on

entrepreneurship and self-employment. In the former case, each increment to the

number of layers in the firm significantly reduces the odds of entrepreneurship. In the

latter case, movement between zero and any other number of layers reduces self-

employment, but there is no effect of any increment beyond zero layers. The effect of

layers on movement to other incumbent firms, while statistically significant on account

of the large sample size, is small: in the largest estimated effect, when moving from

zero to two layers of management, the relative odds of job switching decline by less

than 8 percent.

9 Supervisors are about 25 percent more likely than production workers to become self-employed. Senior staff is 50 percent and CEOs are less than ten percent more likely to do so.

13

0

1

2

3

4

0

1

2

3

4

Od

ds

Rat

io

Rank

Entrepreneurship

Self-employmentJob switching

Figure 1. Odds Ratios of Mobility, by destination and rank, with 95% confidence intervals. From model A of Table 7.

0

1

0

0.2

0.4

0.6

0.8

1

1.2

Od

ds

Rat

io

Firm Size (no. of employees)

Entrepreneurship Self-employment

Job switching

Figure 2. Odds Ratios of Mobility, by destination and firm size, with 95% confidence intervals. From model A of Table 7.

14

Although management layers have a profound effect on the likelihood of transitions

into entrepreneurship and self-employment, and the number of layers is strongly cor-

related with firm size, adding layers to the regressions explains only a modest fraction

of the small firm effect. All six coefficients for firm size in columns (4) and (5) are

smaller than their counterparts in columns (1) and (2), consistent with our conjecture

that management layers may explain part of the small firm effect. However, the reduc-

tions are rather modest, ranging from 13 to 20 percent for entrepreneurship, 4 to 10

percent for self-employment, and from 2 to 17 percent for job switching (see Figure 4).

10 Thus, although we have found that layers of management have a large effect on the

likelihood of a transition to entrepreneurship and to a somewhat lesser extent on the

likelihood of a transition to self-employment, that layers and firm size are positively

correlated, and that firm size is negatively correlated with mobility, layers explain on

average only about 10 percent of the estimated firm size effect.

Because self-employment and entrepreneurship are likely strong substitutes among

those considering separation from their employer, it is useful to verify the robustness

of the results in Table 7 by restricting the choice set. We do so by examining the odds

of self-employment and entrepreneurship relative to job switching after restricting the

sample to movers. Table 8 reports estimates of multinomial logit estimation on this

10 The largest reduction in coefficient size for job switching is based on a small initial effect: the odds ratio of switching to an incumbent employer for employees in size class 50–100 (relative to less than 50 employees) increases from 0.938 to 0.949 after the addition of controls for lay-ers.

0.5

1.1

0.5

0.6

0.7

0.8

0.9

1

1.1

1 2 3 1 2 3 1 2 3

Od

ds

Ra

tio

Layers of Management

Entrepreneurship Self-employment

Job switching

Figure 3. Odds Ratios of Mobility, by destination and number of layers in employee’s firm, with 95% confidence intervals. From model B of Table 7

15

0

5

10

15

20

25

30

50-100 100-500 500-1500 >1500

Per

ecen

tage

red

uct

ion

in r

egre

ssio

n c

oef

fici

ent

Firm Size Class (number of employees)

Entrepreneurship Self-Employment

restricted sample; the baseline category is moving to an incumbent firm.11 The results

clearly demonstrate the small firm effect on business creation: among movers, and

conditional on rank (model A) and on both rank and layers (model B), the likelihood of

business creation instead of job switching declines markedly as firm size increases.

Model B shows that, conditional on size and rank, the likelihood of business creation

also declines markedly as the number of layers increases.

11 If we were to rerun Table 7 using moving to another firm as the baseline, it would of course be possible to formally test the IIA assumption using the Hausman-McFadden (1984) test. If the results in Table 7 are robust, we should find that the difference between any coefficient on en-trepreneurship or self-employment and the corresponding coefficient on self-employment in Table 7 is similar to the corresponding coefficient on entrepreneurship or self-employment in Table 8. The results meet this expectation.

0

5

10

15

20

25

50-100 100-500 500-1500 >1500

Per

ecen

tage

red

uct

ion

in r

egre

ssio

n c

oef

fici

ent

Firm Size Class (number of employees)

Entrepreneurship Self-employment Job switching

Figure 4. Percentage reduction in regression coefficients on firm size obtained upon adding controls for layers of management.

Figure 5. Percentage reduction in regression coefficients on firm size obtained upon adding controls for layers of management. Movers only.

16

Table 8 Transitions among movers. Multinomial logit estimation

Model A Model B

Entrepreneurship Self-employment Entrepreneurship Self-employment

CEOs and Directors 1.220*** 0.244*** 1.293*** 0.267*** (0.071) (0.062) (0.072) (0.062) Senior staff 0.846*** 0.154*** 0.895*** 0.168*** (0.055) (0.037) (0.055) (0.038) Supervisors 0.734*** 0.220*** 0.762*** 0.228*** (0.039) (0.020) (0.039) (0.020) Firm layers: 1 -0.174*** -0.164*** (0.061) (0.033) Firm layers: 2 -0.285*** -0.191*** (0.061) (0.033) Firm layers: 3 -0.426*** -0.176*** (0.064) (0.034) Size 50-100 -0.487*** -0.329*** -0.394*** -0.298*** (0.048) (0.027) (0.050) (0.028) Size 100-500 -0.741*** -0.531*** -0.610*** -0.495*** (0.038) (0.021) (0.043) (0.024) Size 500-1500 -0.664*** -0.589*** -0.512*** -0.552*** (0.044) (0.026) (0.049) (0.029) Size >1500 -0.746*** -0.567*** -0.579*** -0.527*** (0.043) (0.023) (0.050) (0.028) Age (years) 0.204*** 0.176*** 0.204*** 0.176*** (0.012) (0.006) (0.012) (0.006) Age squared -0.002*** -0.002*** -0.002*** -0.002*** (0.000) (0.000) (0.000) (0.000) Female = 1 -0.826*** -0.545*** -0.822*** -0.543*** (0.039) (0.019) (0.039) (0.019) Education -0.027* 0.006 -0.027* 0.006 (0.015) (0.008) (0.015) (0.008) Log(wage) 0.676*** -0.262*** 0.680*** -0.261*** (0.031) (0.011) (0.031) (0.011) Tenure (years) 0.057*** 0.031*** 0.058*** 0.031*** (0.010) (0.006) (0.010) (0.006) Tenure squared -0.002*** -0.002*** -0.002*** -0.002*** (0.001) (0.000) (0.001) (0.000) Constant -13.766*** -6.001*** -13.565*** -5.867*** (0.288) (0.128) (0.291) (0.130)

N 828,683 828,683 828,683 828,683

Standard errors in parentheses. . *** p

17

with more than zero. Finally, as Figure 5 illustrates, the fraction of the small firm effect

that is explained by layers is somewhat larger when the sample is restricted to movers

than it is in the full sample: Adding layers to the regression induces an average 22 per-

cent decline in the coefficients on size for entry into entrepreneurship, although the

decline remains less than 10 percent for self-employment.

4. Why Hierarchies Matter

In the previous section we found that, conditional on firm size, more layers of man-

agement are associated with less mobility of all kinds. While the magnitude of the im-

pact on movements to other employers is small, more layers markedly suppress rates

of business creation by employees. The contrast between the effects of layers on busi-

ness creation and job switching indicate that the reduced mobility associated with

more layers is not just an artifact of the general correlation between larger firms (with

more layers) and turnover in general. In this section, we consider several candidate

explanations for this finding, and provide some evidence concerning each one.

4.1. Bureaucracy and preferences

There is an extensive literature documenting that some individuals declare, often at an

early stage in their career, a preference to create and operate their own business, and

that such declarations predict entry into entrepreneurship (Hamilton, 2000; Halaby,

2003; Benz and Frey, 2008; Åstebro and Thompson, 2010). Halaby (2003) and Søren-

sen (2007) show that the offspring of self-employed parents, who have on average

much greater entrepreneurial aspirations than the offspring of wage earners, are more

likely to work in smaller and presumably less bureaucratic organizations. Benz and

Frey (2008) have documented that both entrepreneurs and employees of smaller firms

report greater levels of job satisfaction, and that in both cases satisfaction is associated

with having greater autonomy and more rewarding work content. There seems, there-

fore to be an association between the preferences of people who found businesses and

employees of small organizations that might induce a small firm effect.12

Elfenbein et al. (2010) claim that, if preference sorting is the only reason we observe a

small firm effect, then the size of employers should be unrelated to subsequent per-

formance in entrepreneurship and self-employment. By extension, layers of manage-

ment should not predict entrepreneurial performance.13 However, this argument

12 Offsetting this, however, is the likelihood that small firm employment is a substitute for en-trepreneurship among certain types of individuals.

13 Elfenbein et al. report that entrepreneurs previously employed in small firms earn less than those employed by the largest firms, but attribute this to the lower opportunity cost of entre-preneurship among the former group. Once they control for prior wages, they find a 23 percent premium to entrepreneurial earnings among those that left the smallest firms relative to those that left the largest. However, the effect is non-monotonic, with the largest premium occurring

18

strikes us as incomplete. Agents create and then continue to operate businesses if the

total compensation expected from the business exceeds the foregone wage. However,

total compensation from entrepreneurship and self-employment includes a non-

pecuniary component among those with a preference for entrepreneurship. Because

people with such preferences are presumably overrepresented in less-bureaucratic

firms, we may anticipate either or both of (i) a positive relationship between the num-

ber of management layers and subsequent monetary performance in entrepreneur-

ship, and (ii) a negative relationship between the number of management layers and

the likelihood of business survival.

Which of these two possible consequences of preferences matters more depends in

large part on the ability of agents to forecast business earnings. If forecast errors are

large, so entry into business is predicated mostly on the population mean of business

earnings, then there is no mechanism by which agents with a preference for entrepre-

neurship can choose to accept lower income in exchange for the non-pecuniary payoff

that business ownership entails. In this case, monetary business income will not de-

pend upon preferences. However, agents with a preference for entrepreneurship are

more likely to continue operating a business that provides poor financial returns, so

management layers and business survival will be negatively correlated. If, in contrast,

forecast errors are small, agents with a preference for entrepreneurship will elect to

create businesses that are known in advance to offer relatively poor financial returns.

Selection effects imply that they will earn less on average and, more precisely, that

they are more likely to create a business that pays less than they earned in wage em-

ployment. However, because they are on average of worse initial quality, businesses

created by agents with a preference for entrepreneurship do not necessarily survive

longer than those created by agents without such preferences.

To test the thesis of preference sorting, we regress entrepreneurial and self-

employment earnings and survival rates on the size of, and the number of manage-

ment layers in, the business owner’s previous employer. We report the results of these

regressions, separately for entrepreneurs and the self-employed, in Table 9. There is a

positive relationship between the number of layers in the previous employer and an

agent's earnings from entrepreneurship, although this relationship is not apparent for

self-employment. Furthermore, businesses created by agents that left a firm with no

layers of management have higher survival rates, results that hold for both entrepre-

neurship and self-employment. There are no clear patterns between performance and

firm size. The result provide evidence for the existence of both channels through which

preference sorting may affect firm performance, and these effects are more clearly

associated with the hierarchical structure of the firm than they are with firm size.

among entrepreneurs previously employed in medium-sized firms.

19

Table 9 Performance of entrepreneurs, two years after business creation

OLS Regressions

Log(Total income) Survival

Entrepreneurs Self-employed Entrepreneurs Self-employed (1) (2) (3) (4)

CEOs and Directors 0.032 -0.013 -0.030* -0.069*** (0.026) (0.027) (0.018) (0.026) Managers -0.003 -0.032* -0.019 -0.060*** (0.021) (0.016) (0.014) (0.016) Supervisors 0.010 -0.014 0.008 -0.022** (0.015) (0.009) (0.011) (0.009) Firm layers: 1 0.043* 0.004 -0.042** -0.029** (0.024) (0.014) (0.017) (0.014) Firm layers: 2 0.074*** -0.008 -0.049*** -0.036*** (0.024) (0.014) (0.017) (0.014) Firm layers: 3 0.069*** -0.002 -0.047*** -0.038*** (0.025) (0.015) (0.018) (0.014) Size 50-100 -0.018 -0.001 0.016 -0.008 (0.020) (0.012) (0.014) (0.012) Size 100-500 -0.026 0.011 0.021* 0.010 (0.017) (0.011) (0.012) (0.010) Size 500-1500 -0.006 0.018 0.031** -0.010 (0.020) (0.013) (0.014) (0.012) Size 1500> -0.008 0.020 0.020 -0.033*** (0.019) (0.012) (0.014) (0.012) Log(wage) -0.019*** -0.025*** 0.007** -0.002 (0.005) (0.002) (0.003) (0.002) Tenure (years) 0.000*** 0.000*** -0.000*** 0.000 (0.000) (0.000) (0.000) (0.000) Tenure squared -0.005 -0.025*** -0.018 -0.021*** (0.016) (0.008) (0.011) (0.008) Observations 5,769 19,276 5,769 19,276 R squared 0.72 0.76 0.71 0.25

Standard errors in parentheses. *** p

20

4.2 Information processing

Caliendo and Rossi-Hansberg (2012) develop a general equilibrium model of manage-

ment layers in which output is secured by solving problems, and worker knowledge

and time are key inputs into the production process. Workers on the shop floor (their

context is manufacturing) produce by solving problems that present themselves. If

they have the knowledge, they solve the problem themselves, which yields some out-

put. If they do not have the knowledge they pass the problem up to their supervisor,

who may either solve the problem or pass it up to the next layer of management. Em-

ployees are constrained not only by their knowledge, but also by time. Thus, a supervi-

sor in charge of too many workers, or of workers with too little knowledge, will not be

able to address all the problems that come up the line. Caliendo and Rossi-Hansberg

examine the optimal number of management layers, and size and composition of each

layer, and how these change as a firm experiences an increase in demand.

Firm heterogeneity is an important part of the model, and firms may respond to an

increase in demand in either of two ways. First, they may increase the size and change

the composition of each layer, while holding the number of layers constant. Alterna-

tively, they may add a layer of management, which in turn induces changes in the size

and composition of existing layers. Caliendo and Rossi-Hansberg show that when a

firm expands layers, it increases the number of hours worked at each pre-existing lay-

er. The additional layers enable workers to pass more problems up the line. They

therefore need less knowledge and the average wage declines. Symmetrically, elimina-

tion of layers reduces layer size and raises the average wage within each surviving

layer. In contrast, if a firm expands without increasing layers it must pay higher wages

at each level because time constraints on upper management force workers in lower

layers to solve more of their own problems. Caliendo, Monte and Rossi-Hansberg

(2012) exhaustively examine these predictions of the model using a panel of observa-

tions on the large majority of French manufacturing firms, and find no instance in

which the evidence contradicts the predictions. Tåg (2013) replicates their study and

concludes that the predictions of Caliendo and Rossi-Hansberg (2012) also hold for

Swedish manufacturing firms.

The theory predicts that conditional on firm size, firms with fewer layers employ

workers with the ability to solve a broader range of problems. Consistent with Lazear’s

(2005) Jack-of-all-trades theory, such workers are more likely to enter entrepreneur-

ship and self-employment, and they are likely to perform better when they do. Howev-

er, workers with broader skills are paid a higher wage, which reduces mobility. If we

condition on an individual’s wage, however, employees in firms of a given size but with

fewer layers should have a higher propensity to enter entrepreneurship and self-

employment and be more successful when they do. Similarly, employees of larger

firms with a given number of layers should, conditional on their wage, be more likely

to create a business and be more successful when they do.

21

The evidence from analyses we have already reported is not especially favorable to

these predictions. In Table 8, which estimated business creation rates after controlling

for prior earnings, we saw that while more layers reduce business creation, larger

firms were associated with less entry. The effect of layers is consistent with the model,

but the effect of firm size is not. In the performance regressions of Table 9, we found

that more layers are associated with higher entrepreneurial earnings and lower sur-

vival rates. Both these associations, we concluded, were consistent with preference

sorting; the effect of layers on business earnings is not what the information pro-

cessing model would lead us to expect.

Of course, preference sorting and information processing are not mutually exclusive

theories, so it is quite possible that evidence for the latter theory is confounded by

preference sorting. We can conduct a perhaps sharper test of the information pro-

cessing story. So far, we have controlled in our multinomial logit regressions for rank,

the number of layers and firm size. Of particular note here is the positive effect we

found of higher rank on the likelihood of business creation, and especially of entrepre-

neurship. However, the meaning of a particular rank depends on the hierarchical

structure of the firm. The information processing story implies that the more layers

there are in a firm the greater the difference between the breadth of knowledge of em-

ployees in the highest and lowest ranks. As a result, the likelihood of business creation

of any rank relative to the lowest rank in the firm should be greater the more ranks

that lie between them.

Evidence on this prediction is provided in Table 10, which estimates the effect of rank

on the likelihood of business creation separately for firms with different numbers of

layers. Consistent with the information processing story, we find first that for any giv-

en number of layers in a firm the probability of entering entrepreneurship and self-

employment is greater the higher the relative position. Moreover, the coefficients for

entrepreneurship and self-employment on the top layers are larger the more layers

there are in the firm. This is true for employees in the top ranks as we move from one-

layer to two-layer and then to three-layer firms; and it is also true for employees in the

second highest rank as we move from two to three layers in the firm. These findings

hold both for entrepreneurship and for self-employment and they suggest, consistent

with Caliendo and Rossi-Hansberg (2012), that the span of knowledge of managers is

related to the number of layers beneath them, rather than the title of their rank.

Table 10 Worker Transitions, Multinomial Logit estimations

THREE LAYERS

TWO LAYERS

ONE LAYER

Entrepreneurship

Self-

employment Other firm Entrepreneurship

Self-

employment Other firm Entrepreneurship

Self-

employment Other firm

Top

1.460*** 0.534*** 0.300***

0.747*** 0.314*** 0.187***

0.516*** -0.017 -0.233***

(0.099) (0.081) (0.014)

(0.090) (0.064) (0.012)

(0.081) (0.052) (0.011)

Second

1.153*** 0.426*** 0.293***

0.565*** 0.255*** 0.068***

(0.074) (0.046) (0.008)

(0.065) (0.037) (0.007)

Third

0.865*** 0.249*** 0.035***

(0.055) (0.027) (0.004)

Size 50-100

-0.311*** -0.192*** -0.035***

-0.394*** -0.343*** -0.055***

-0.555*** -0.469*** -0.086***

(0.081) (0.048) (0.009)

(0.079) (0.046) (0.008)

(0.122) (0.065) (0.012)

Size 100-500

-0.641*** -0.514*** -0.097***

-0.624*** -0.575*** -0.182***

-0.909*** -0.764*** -0.040***

(0.067) (0.041) (0.008)

(0.065) (0.039) (0.007)

(0.163) (0.083) (0.013)

Size 500-1500

-0.667*** -0.708*** -0.248***

-0.792*** -0.754*** -0.222***

-1.275*** -1.136*** -0.219***

(0.071) (0.043) (0.008)

(0.090) (0.054) (0.009)

(0.257) (0.166) (0.022)

Size 1500>

-1.243*** -1.024*** -0.513***

-1.074*** -1.107*** -0.675***

-0.708*** -1.191*** -0.753***

(0.074) (0.042) (0.008)

(0.091) (0.056) (0.010)

(0.191) (0.129) (0.022)

Log(wage)

0.467*** -0.784*** -0.527***

0.589*** -0.738*** -0.497***

0.514*** -0.760*** -0.575***

(0.049) (0.014) (0.003)

(0.066) (0.023) (0.005)

(0.091) (0.028) (0.008)

(0.001) (0.001) (0.000)

(0.001) (0.001) (0.000)

(0.002) (0.001) (0.000)

Constant

-12.625*** -2.989*** 3.428***

-13.137*** -3.345*** 2.765***

-13.776*** -4.065*** 2.917***

(0.425) (0.179) (0.027) (0.573) (0.263) (0.044) (0.864) (0.366) (0.068)

Observations

4,561,415 4,561,415 4,561,415

1,525,459 1,525,459 1,525,459 548,383 548,383 548,383

Standard errors in parentheses. *** p

4.3 Career concerns

A firm may develop a hierarchy to create incentives for employees to exert effort. In par-

ticular, tournament theory (e.g., Lazear and Rosen, 1981; Lazear and Shaw, 2007) explains

that wages are generally associated with jobs rather than with individuals, and that pro-

motion is the reward for employees whose performance exceeds that of their peers. Con-

ditional on size, a firm with more layers offers greater prospects of promotion, so employ-

ees are more likely to remain with the firm. The prospect of promotion also serves as a

form of deferred compensation, which adds a further incentive to remain with a firm.

Thus, the concerns that individuals have for their career prospects may well lie behind the

negative association we have observed between the number of layers and mobility of all

kinds.

Unless employees care about status, the incentives that layers provide to remain with the

firm depend on how much pay increases at the time of promotion. We would therefore

expect a larger within-firm difference in wages across ranks to increase the incentive to

stay in the organization. In model A of Table 11, we test this notion by estimating transi-

tion probabilities as a function of, inter alia, the standard deviation of wages within the

firm. The sample is restricted to firms with at least one layer of management. The regres-

sion also controls for an agent’s own wage and the average wage in the firm, the latter

because we already know that it is correlated with the number of layers. The results are

not what the career concerns model would lead us to expect: the greater the wage disper-

sion the higher the probability of exit to entrepreneurship, self-employment and other

firms. Our results for mobility are the same as those reported by Lazear and Shaw (2008)

across eight countries.

The standard deviation of wages reflects not only the increments to the wage that are se-

cured by promotion to a higher rank but also within-rank dispersion. Although this has

been the usual way in which a firm’s wage policy has been measured, it might be a poor

measure for tests of the career concerns model. In Model B of Table 11 we replace the

standard deviation with the firm’s average between-rank change of wage, estimated

through a linear regression of wage on rank across all employees within the firm. This

measure might more adequately reflect the incentive effects of promotion. However we

continue to find no support for the career concerns model. Although the coefficient on this

Table 11 Worker Transitions. Multinomial Logit Estimation

Model A Model B Model C

VARIABLES Entrepre-neurship

Self-employment

Other firm

Entrepre-neurship

Self-employment

Other firm

Entrepre-neurship

Self-employment

Other firm

CEOs and directors 1.278*** 0.445*** 0.295*** 1.254*** 0.454*** 0.306*** 1.257*** 0.453*** 0.306***

(0.097) (0.086) (0.014) (0.097) (0.086) (0.014) (0.097) (0.086) (0.014)

Senior staff 1.053*** 0.415*** 0.304*** 1.041*** 0.424*** 0.309*** 1.042*** 0.424*** 0.309***

(0.063) (0.041) (0.007) (0.063) (0.041) (0.007) (0.063) (0.041) (0.007)

Supervisors 0.821*** 0.205*** 0.058*** 0.820*** 0.212*** 0.062*** 0.821*** 0.212*** 0.061***

(0.047) (0.024) (0.004) (0.047) (0.024) (0.004) (0.047) (0.024) (0.004)

Firm layers: 2 -0.040 0.033 -0.059*** -0.044 0.052 -0.034*** -0.040 0.052 -0.031***

(0.073) (0.041) (0.007) (0.074) (0.042) (0.007) (0.074) (0.042) (0.007)

Firm layers: 3 -0.138* 0.059 -0.035*** -0.135* 0.087** -0.005 -0.131* 0.086** -0.003

(0.072) (0.041) (0.007) (0.073) (0.041) (0.007) (0.073) (0.041) (0.007)

Log(wage) 0.569*** -0.768*** -0.510*** 0.608*** -0.770*** -0.509*** 0.607*** -0.770*** -0.509***

(0.045) (0.013) (0.003) (0.045) (0.013) (0.003) (0.045) (0.013) (0.003)

Firm mean log(wage) -0.097 0.254*** -0.081*** -0.190** 0.136*** -0.152*** -0.193** 0.136*** -0.149***

(0.093) (0.051) (0.008) (0.091) (0.048) (0.008) (0.091) (0.048) (0.008)

Firm log(wage) standard deviation

0.514*** 0.565*** 0.347***

(0.139) (0.081) (0.014)

Wage slope

-0.188* 0.067 0.137***

(0.100) (0.058) (0.009)

Relative wage slope

-0.059 0.026 0.066***

(0.039) (0.022) (0.004)

Observations 5,952,282 5,952,282 5,952,282 5,952,282 5,952,282 5,952,282 5,952,282 5,952,282 5,952,282

Standard errors in parentheses, *** p

wage slope variable has the expected sign for entrepreneurship, it continues to have the

wrong sign for self-employment and job switching.

Perhaps we are using the wrong measure of wage dispersion to proxy incentives to stay

and work. The appropriate measure of dispersion as an incentive for an agent to stay with

his current employer may be one that is calculated relative to the dispersion that is found

at other firms that might employ the agent. We explore this possibility next. We use our

measure of the wage slope across ranks, but now we compare it with the slopes of firms

operating in the same industry and year.14 The results, shown as Model C of Table 11 again

fail to offer support for the career concerns.

4.4 Stars and Misfits

Throughout much of our analysis so far, we have seen different results for entry into self-

employment and entrepreneurship. This is particularly so with the various measures of

earnings that we have included: the agent’s own wage is positively associated with entry

into entrepreneurship but negatively with self-employment (Tables 7 and 8); the firm’s

mean wage and between-rank wage slope also have effects of opposite signs (although not

always statistically significant) on entrepreneurship and self-employment (Table 11); and

the number of layers in the firm is positively related to subsequent entrepreneurial earn-

ings but not to self-employment earnings (Table 9). These contrasting impacts suggest

that entrepreneurship and self-employment may be chosen by qualitatively different em-

ployees.

In a recent paper, Åstebro, Chen and Thompson (2011) developed and tested a model in

which workers vary in the degree to which they are well matched to their job. Those that

are poorly matched earn less than they might expect in business creation, and so they

leave. Although workers are treated symmetrically in their model, Åstebro et al. show that

mismatches are more common in the tails of the ability and earnings distributions. In the

lower tail, agents with low ability are more likely to enter self-employment, while in the

upper tail, high-ability agents are more likely to become entrepreneurs. If one is willing to

suppose that low-ability workers creating a business are likely to become self-employed

while their high-ability counterparts are likely to establish limited liability companies, the

mismatching model readily accounts for the contrasting effects of own wage on business

14 The wage slope is based on a regression of labor income on the management layer in a firm and is normalized such that it is relative to the industry-year mean slope. It is positive if the wage slope is higher than industry-year mean and negative otherwise. Because the slope is meaningless in firms with zero layers of management, we drop these firms. We also restrict attention to firms with more than 30 employees to have large enough samples for estimating the wage-rank slope for each firm individually.

26

creation. Among the set of agents at the lower end of the distribution a lower wage is asso-

ciated on average with less ability and a greater likelihood of suffering a job mismatch.

Thus, own wage is negatively correlated with entry into self-employment. In contrast, a

higher wage among agents at the top of the earnings distribution is associated on average

with greater ability, and an increased chance of a mismatch, so own wage is positively

correlated with entrepreneurship.

It is less apparent that the mismatching model can explain the effects of management lay-

ers on mobility. One possible mechanism, however, is that layers create a hierarchy that

induces a functional separation between agents of high and low ability, and thereby re-

duces mismatching problems.15 However, our view is that the main impact of stars and

misfits is to insert some anomalies in our empirical tests of preference sorting, infor-

mation processing, and career concerns as potential explanations for the effects of layers.

To see whether mismatching might be influencing our results, we examine where in the

wage distribution the probability of exit is the greatest. To do so we create a dummy for

each decile of the wage distribution and examine the effects of these dummies on exit

probabilities, leaving the 40th to the 60th percentile of the wage distribution as the omitted

category. We include each employee’s own wage in the regression to absorb effects on

wage that are not related to within-firm relative ability. The results, which are reported in

Table 12 are quite striking. The probability of remaining in the firm is clearly highest

among those with middling wages while the probability of exit is remarkably higher

among those in the bottom and top quintiles of the earnings distribution. The bottom

quintile is much more likely than those with middling earnings to enter self-employment

or to join another firm, and they are also somewhat more likely to become entrepreneurs.

In contrast, the top earners are much more likely to become entrepreneurs, more likely to

switch jobs, but less likely to become self-employed. These patterns exhibit a noteworthy

consistency with the mismatching model. However, they appear to provide no explanation

for the effect of layers, the coefficients on which are all but identical to those obtained in

our previous regressions.

5. Conclusions

In this paper, we asked whether small firms are more frequent incubators of entrepre-

neurs because they tend to be less hierarchical. We found that hierarchy, at least as we

have measured it in terms of the number of layers of management, is indeed less prevalent

in small firms and is associated with more frequent transitions of employees into self-

employment and entrepreneurship. However, we also found that hierarchy explains no

15 In Åstebro et al. mismatching arises because agents of differing ability may become co-workers and the production function exhibits complementarity in worker abilities.

27

Table 12

Worker Transitions. Multinomial Logit Estimation

Entrepreneurship Self-employment Other firm

CEOs and Directors 1.007*** 0.011 -0.187*** (0.068) (0.061) (0.012) Senior staff 0.867*** 0.370*** 0.084*** (0.056) (0.039) (0.007) Supervisors 0.693*** 0.334*** 0.009** (0.042) (0.021) (0.004) Log(wage) 0.183*** -0.367*** -0.285*** (0.063) (0.021) (0.005) Firm layers: 1 -0.239*** -0.185*** -0.029*** (0.061) (0.032) (0.007) Firm layers: 2 -0.325*** -0.215*** -0.057*** (0.061) (0.032) (0.007) Firm layers: 3 -0.448*** -0.185*** -0.034***

(0.063) (0.034) (0.007) 0-10th percentile 0.351*** 1.001*** 0.696*** (0.106) (0.038) (0.007) 10-20th percentile 0.013 0.692*** 0.363*** (0.077) (0.027) (0.005) 20-30th percentile -0.099 0.285*** 0.120*** (0.071) (0.028) (0.005) 60-70th percentile 0.005 -0.030 0.019*** (0.056) (0.029) (0.005) 70-80th percentile 0.264*** -0.143*** 0.082*** (0.053) (0.031) (0.005) 80-90th percentile 0.426*** -0.097*** 0.179*** (0.055) (0.032) (0.006) +90th percentile 0.757*** 0.040 0.341*** (0.071) (0.037) (0.007) N 6,865,026 6,865,026 6,865,026

Standard errors in parentheses. *** p

28

ployment in firms with fewer management layers (preference sorting). Second, employees

in firms with fewer layers have a broader range of skills, and this makes them more fit for

entrepreneurship (ability breadth). Third, multiple layers of management offer promotion

opportunities that do not exist in firms without a hierarchy, thereby suppressing mobility

(career concerns). Finally, skill mismatching is less problematic in firms with a more hier-

archical structure. We found evidence for preference sorting, ability breadth and skill

mismatching, but none for career concerns. However, there remained throughout a sub-

stantial unexplained component of the effect of management layers on business creation

rates.

Our analysis is based on the Swedish matched employer-employee dataset. Conventional

wisdom has it that, along with other Scandinavian countries, Sweden’s labor market is

atypical. This conventional wisdom is now dated, 16 and most employment data reveal that

Sweden is comparable to other OECD countries. While labor mobility remains lower than

average, its wage structure is now much like other high-income countries. In fact, Lazear

and Shaw (2008), show that across several measures such as wage dispersion within

firms, the variance of wage growth rates within firms, and even overall wage dispersion,

Sweden is not remarkably different from seven other countries they study, including the

U.S.A. The environment for new firm formation in Sweden is further not markedly differ-

ent from other countries such as the U.S.A., Brazil or Denmark (Andersson and Klepper,

2013). We thus see no reason why results found in this paper would not be replicated

elsewhere.

References

Anderson, Patricia M., and Bruce D. Mayer (1994): “The extent and consequences of job turno-ver.” Brookings Papers on Economic Activity: Microeconomics, 177–236.

Andersson, Martin, and Steven Klepper (2013): ”Characteristics and performance of new firms and spinoffs in Sweden.” Industrial and Corporate Change, 22(1): 245–80.

Åstebro, Thomas, Jing Chen, and Peter Thompson (2011): “Stars and misfits: Self-employment and labor market frictions.” Management Science, 57(11): 1999–2017.

16 For the majority of the 20th century, annual average wage increases were set by “collective bar-gaining” between the three major labor unions and the three major employers, respectively. How-ever, between 1970 and 1990 centralized control over wages was progressively dismantled, and it was abandoned entirely in 1997. Further deregulations of the Swedish labor market in the early 1990s gave rise to a dual labor market consisting of temporary and permanently employed work-ers. These changes induced an increase in the Gini coefficient for income from 0.199 in 1981 to 0.332 in 2007 (http://www.scb.se/Pages/TableAndChart____163550.aspx). The 2007 figure places Sweden at the OECD average (The Economist, 2013). The percentage of temporary employees in the private sector rose from 8.8 percent in 1997 to 13.1 percent in 2008 (Sveriges Officiella Statistik, 2001; 2012).

29

Åstebro, Thomas, and Peter Thompson (2010): “Entrepreneurs: Jacks of all trades or ho-bos?” Research Policy, 40(5): 637–649.

Benz, Matthias and Bruno S. Frey (2008): “Being independent is a great thing: Subjective eval-uations of self-employment and hierarchy.” Economica, 75(298):362–383.

Caliendo, Lorenzo, Ferdinando Monte, and Esteban Rossi-Hansberg (2012): “The anatomy of French production hierarchies.” Working paper, Yale University.

Caliendo, Lorenzo, and Esteban Rossi-Hansberg (2012): “The impact of trade on organization and productivity.” Quarterly Journal of Economics, 127(3):1393–1467.

Chen, Jing (2012): “The small firm effect and the quality of entrepreneurs.” Working paper, Copenhagen Business School.

Dobrev, Stanislav D. and William P. Barnett (2005): “Organizational roles and transition to entrepreneurship.” Academy of Management Journal, 48(3): 433–449.

Elfenbein, Daniel W., Barton H. Hamilton, and Todd R. Zenger (2010): “The small firm effect and the entrepreneurial spawning of scientists and engineers.” Management Science, 56(4):659–681.

Frederiksen, Anders (2006): “Gender differences in job separation rates and employment

stability: New evidence from employer-employee data.” Institute for the Study of

Labor (IZA) Discussion Paper No. 2147.

Gompers, Paul, Josh. Lerner, and David Scharfstein (2005): “Entrepreneurial spawning: Public corporations and the genesis of new ventures, 1986 to 1999.” Journal of Finance, 60(2): 577–614.

Halaby, Charles N. (2003): “Where job values come from. Family and schooling background, cognitive ability, and gender.” American Sociological Review, 68; 251–278.

Haltiwanger, John, Ron S. Jarmin, and Javier Miranda (2011): “Who creates jobs? Small vs. large vs. young.” Working paper, University of Maryland.

Hamilton, Barton (2000): “Does entrepreneurship pay? An empirical analysis of the returns to self-employment.” Journal of Political Economy, 108(3): 604–631.

Hausman, Jerry, and Daniel McFadden (1984): “Specification tests for the multinomial logit model." Econometrica, 52(5): 1219–1240.

Henrekson, Magnus and Tino Sanandaji, 2013, "Billionaire Entrepreneurs: A Systematic Analy-sis", IFN Working Paper No. 959

Lazear, Edward P., and Sherwin Rosen (1981): “Rank-order tournaments as optimum labor contracts.” Journal of Political Economy, 89(5): 841–64.

Lazear, Edward P. (2005):”Entrepreneurship.” Journal of Labor Economics, 23(4):649-680.

Lazear, Edward P. and Kathryn L. Shaw (2007). "Personnel economics: The economist's view of human resources." Journal of Economic Perspectives, 21(4): 91–114.

Lazear, Edward P. and Kathryn L. Shaw, (2008)” “Introduction: Wage structure, raises and mobility: International comparisons of the structure of wages within and across firms.” In E. Lazear and K. Shaw eds., The Structure of Wages: An International Comparison, Uni-versity of Chicago Press.

Sørensen, Jesper (2007): ‘Bureaucracy and entrepreneurship: Workplace effects on entrepre-

30

neurial entry.” Administrative Science Quarterly, 52; 387–412.

Sveriges Officiella Statistik (2001): Statistiska Meddelanden, AM63 SM 0102, Statistiska Centralbyrån.

Sveriges Officiella Statistik (2012): Statistiska Meddelanden, AM63 SM 1204, Statistiska Centralbyrån.

Tåg, Joacim (2013): “Production Hierarchies in Sweden”, IFN Working Paper No. 963

The Economist (2013): “The ins and the outs. Immigration and growing inequality are making the Nordics less homogeneous.” The Economist, February 2nd, 2013.

Wagner, Joachim (2004): “Are young and small firms hothouses for nascent entrepreneurs?” Applied Economics Quarterly, 50; 379–391.